Actually, if a crackpot can faithfully camouflage their output then they are probably not crackpotty enough to deserve deletion without any consideration at all

Actually, if a crackpot can faithfully camouflage their output then they are probably not crackpotty enough to deserve deletion without any consideration at all. Imagine, you have succesfully proved P eq/neq NP... but you can't be bothered to learn LaTeX or convince someone who does to typeset your work? Unfreeking likely. The vast bulk of the junk can be disregarded by simple heuristics because the crackpots are unwilling or unable to meet even a very low bar. Telling them precisely where the bar is will only get a small amount past it.